Get your first YAM actions

git clone https://github.com/reactor-team/reactor-cookbook.git
cd reactor-cookbook
cd robotics/sim/notebooks
uv sync --frozen --python 3.12
export REACTOR_API_KEY='your-api-key'

uv run --frozen python first_dreamzero_yam_actions.py

import asyncio
import time

import numpy as np
from reactor_robotics.session import ReactorSession

TRACKS = ('top', 'left', 'right')
STATE = {'left_joint_pos': [0.0, 0.0, 0.0, 0.0, 0.0, 0.0], 'left_gripper_pos': 0.0, 'right_joint_pos': [0.0, 0.0, 0.0, 0.0, 0.0, 0.0], 'right_gripper_pos': 0.0}


async def main():
    session = ReactorSession("reactor/dreamzero-yam-molmoact2", fps=15, frame_size=(176, 320))
    try:
        await session.connect(
            TRACKS,
            subscribe=("action_chunk", "command_error", "episode_started", "prompt_accepted"),
            ready_timeout_s=900,
        )
        # Synthetic fixtures: never execute their predicted actions on hardware.
        rng = np.random.default_rng(7)
        frames = {name: rng.integers(0, 256, (176, 320, 3), dtype=np.uint8)
                  for name in TRACKS}
        for field, value in STATE.items():
            await session.send("set_" + field, {field: value})
        session.set_frames(frames)
        await asyncio.sleep(0.5)
        started = time.perf_counter()
        await session.send("set_prompt", {"prompt": "put the block in the bin"})
        reply = await session.next_message("action_chunk", timeout_s=300)
        actions = np.asarray(reply["actions"], dtype=np.float64)
        assert actions.shape == (24, 14), actions.shape
        assert np.isfinite(actions).all()
        assert reply["chunk_index"] == 0, reply["chunk_index"]
        print("actions:", actions.shape)
        print("first row:", actions[0])
        print("obs_seq:", reply["obs_seq"])
        print("model inference seconds:", reply["inference_seconds"])
        print("prompt-to-chunk seconds:", round(time.perf_counter() - started, 3))
    finally:
        await session.close()


asyncio.run(main())